regression diagnostics

    1

    An evidence-based auditor for R and statsmodels regression summaries that checks diagnostics and flags causal overreach.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    regression diagnostics

    Example session with this skill installed

    Check this statsmodels OLS summary for a paper I'm writing:
    Dep. Variable: Y, R-squared: 0.045, Adj. R-squared: 0.012, F-statistic: 1.35, Prob (F-statistic): 0.25. 5 predictors, 30 observations.

    • Read your context and instructions
    • Compiled the regression diagnostics

    I have audited the provided statsmodels OLS summary. Below is the six-dimension diagnostic report based on the provided values.

    Regression Diagnostic Report

    DimensionStatusEvidenceInterpretation
    Sample SizeWARNINGn=30, k=5Fails the 10x rule (n < 50). Model may be overfitted or underpowered.
    Model FitWARNINGF-p=0.25, R²=0.045The overall model is not statistically significant. Interpretation of coefficients is discouraged.
    CoefficientsMISSINGNo SE/P-values providedCannot verify significance or effect sizes for individual predictors.
    CollinearityMISSINGVIF not reportedRisk of multicollinearity cannot be assessed from this summary.
    ResidualsMISSINGNo tests providedLinearity, homoscedasticity, and normality assumptions are unchecked.
    BoundariesPASSNo causal claims madeInput was purely numerical; no unjustified causal language detected.

    Missing Items

    • Coefficient table (Estimates, Standard Errors, p-values, Confidence Intervals).
    • Variance Inflation Factor (VIF) for all 5 predictors.
    • Residual diagnostic tests (e.g., Breusch-Pagan, Shapiro-Wilk).

    Boundary Statement

    Evidence Limitation: With a non-significant F-statistic (0.25), this model fails to reject the null hypothesis. Do not report individual predictors as significant.

    Causal Warning: These results show correlation only. Avoid using terms like "impact," "influence," or "leads to" in your draft.

    Next steps

    1. Re-run the model in your environment to extract the full coefficient table and VIF values.
    2. Perform residual diagnostics to ensure the OLS assumptions are not violated before reporting.
    3. Consider increasing sample size or simplifying the model given the low observations-to-predictor ratio.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Regression analysis is prone to over-interpretation, missing assumptions, and causal fallacies. This skill audits R summary() or Python statsmodels OLS outputs to ensure your statistical reporting is grounded in evidence.

    What it does

    • Data Validation extracts sample sizes and predictor counts to check against heuristic rules like the 10x rule for degrees of freedom.
    • Model Diagnostics evaluates R-squared, Adjusted R-squared, and F-statistics to confirm overall model fit.
    • Coefficient Audit verifies p-values against confidence intervals and flags missing effect size discussions.
    • Assumption Checking scans for collinearity via VIF and identifies missing reports for linearity, independence, homoscedasticity, and normality.
    • Language Scrutiny detects and flags unjustified causal language like "drives" or "causes" in the proposed report context.

    How it works

    1. Provide a raw text block from an R summary(lm_model) or a statsmodels.api.OLS results summary.
    2. The agent extracts every statistical value, mapping them to a structured diagnostic framework.
    3. It applies a six-point check ranging from sample size adequacy to residual assumption reporting.
    4. It generates a standardized report using PASS, WARNING, MISSING, or BLOCKED status for each dimension.

    Frameworks & tools

    Works with standard output from R (base stats) and Python (statsmodels). Can also consume structured JSON for automated pipeline integration.

    Why this beats prompting it yourself

    Generic LLMs often hallucinate p-values or miss the nuance between correlation and causation. This skill enforces a rigid six-dimension check that refuses to "fill in the blanks," ensuring you never report a metric that wasn't actually in your output.

    Use cases

    • Audit a regression summary before submitting a paper to ensure all diagnostic metrics are present.
    • Detect multicollinearity risks by checking VIF values against heuristic thresholds.
    • Sanity-check a colleague's model output to verify if the sample size supports the number of predictors.
    • Review draft report language to remove accidental causal claims before peer review.

    Known limitations

    Does not perform model re-fitting, variable selection, or predictions. It cannot analyze raw datasets (CSV/Parquet) and relies entirely on the provided summary text or JSON.

    Cannot re-fit models or read raw data files. Only audits provided summary text or JSON. Refuses causal inference, variable selection, and predictive tasks.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed today

    • Free with an Agensi account

    Permissions

    Read Files
    Listedtoday

    What's inside

    Frequently Asked Questions

    This skill is designed to work with any AI agent that supports text-based tools or plugins, as long as you can paste your R or statsmodels output into the chat interface. It is optimized for use within research workflows and data science environments.

    No, the skill is a diagnostic auditor, not a data processor. You must provide the text output from an R `summary(lm_model)` or Python `statsmodels` OLS results; the skill then validates that output against statistical best practices.

    It specifically flags "causal overreach" by identifying active verbs like "causes" or "leads to" when the statistical evidence only supports correlation. This helps you avoid common peer-review rejections related to unjustified causal claims.

    The skill uses a rigid six-dimension framework to cross-reference reported metrics, such as ensuring p-values align with confidence intervals. Unlike a general LLM, it is programmed to mark missing data as 'MISSING' rather than hallucinating or guessing the values.

    Yes, while it primarily handles raw text, it also supports structured JSON inputs. This allows you to integrate the diagnostic checks into automated data science pipelines for high-volume model validation.

    New in Research & Analysis